makedbz -i time for history file with 44,759,312 entries
Joe St Sauver
JOE at OREGON.UOREGON.EDU
Fri Jun 8 20:05:46 UTC 2001
After a recent upgrade, it looked as if it was necessary
to rebuild my history dbz files with makedbz -i
The history file itself is 44,759,312 lines long and
so far makedbz -i has been running 4,150 minutes (e.g.,
approach three days) on a gigahertz PIII (it is actually
a dual processor gig PIII, but makedbz isn't threaded,
so one processor sits essentially idle)
(a) is this typical? If so, I think many news admins
may end up just trashing their history files and starting
from scratch rather than being down for three or more days.
The man page might be modified to include a brief summary
of typical run times, if nothing else...
(b) any plans to profile/attack the algorithmic choke
points that are making makedbz -i run for such an
exceptionally long period of time? (the process seems to
be purely CPU bound, at least on my box) In particular,
multithreading the process at a minimum would be a huge
help for folks running in an SMP environment.
More information about the inn-workers